vm_machdep.c revision 83366 
1/*-
2 * Copyright (c) 1982, 1986 The Regents of the University of California.
3 * Copyright (c) 1989, 1990 William Jolitz
4 * Copyright (c) 1994 John Dyson
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * the Systems Programming Group of the University of Utah Computer
9 * Science Department, and William Jolitz.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *	This product includes software developed by the University of
22 *	California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 *    may be used to endorse or promote products derived from this software
25 *    without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
40 *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
41 * $FreeBSD: head/sys/amd64/amd64/vm_machdep.c 83366 2001-09-12 08:38:13Z julian $
42 */
43
44#include "opt_npx.h"
45#ifdef PC98
46#include "opt_pc98.h"
47#endif
48#include "opt_reset.h"
49#include "opt_isa.h"
50#include "opt_kstack_pages.h"
51
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/malloc.h>
55#include <sys/proc.h>
56#include <sys/bio.h>
57#include <sys/buf.h>
58#include <sys/vnode.h>
59#include <sys/vmmeter.h>
60#include <sys/kernel.h>
61#include <sys/ktr.h>
62#include <sys/mutex.h>
63#include <sys/smp.h>
64#include <sys/sysctl.h>
65#include <sys/unistd.h>
66
67#include <machine/cpu.h>
68#include <machine/md_var.h>
69#include <machine/pcb.h>
70#include <machine/pcb_ext.h>
71#include <machine/vm86.h>
72
73#include <vm/vm.h>
74#include <vm/vm_param.h>
75#include <sys/lock.h>
76#include <vm/vm_kern.h>
77#include <vm/vm_page.h>
78#include <vm/vm_map.h>
79#include <vm/vm_extern.h>
80
81#include <sys/user.h>
82
83#ifdef PC98
84#include <pc98/pc98/pc98.h>
85#else
86#include <i386/isa/isa.h>
87#endif
88
89static void	cpu_reset_real __P((void));
90#ifdef SMP
91static void	cpu_reset_proxy __P((void));
92static u_int	cpu_reset_proxyid;
93static volatile u_int	cpu_reset_proxy_active;
94#endif
95extern int	_ucodesel, _udatasel;
96
97/*
98 * quick version of vm_fault
99 */
100int
101vm_fault_quick(v, prot)
102	caddr_t v;
103	int prot;
104{
105	int r;
106
107	if (prot & VM_PROT_WRITE)
108		r = subyte(v, fubyte(v));
109	else
110		r = fubyte(v);
111	return(r);
112}
113
114/*
115 * Finish a fork operation, with process p2 nearly set up.
116 * Copy and update the pcb, set up the stack so that the child
117 * ready to run and return to user mode.
118 */
119void
120cpu_fork(td1, p2, flags)
121	register struct thread *td1;
122	register struct proc *p2;
123	int flags;
124{
125	register struct proc *p1;
126	struct thread *td2;
127	struct pcb *pcb2;
128#ifdef DEV_NPX
129	int savecrit;
130#endif
131
132	p1 = td1->td_proc;
133	td2 = &p2->p_thread;
134	if ((flags & RFPROC) == 0) {
135		if ((flags & RFMEM) == 0) {
136			/* unshare user LDT */
137			struct pcb *pcb1 = td1->td_pcb;
138			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
139			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
140				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
141				if (pcb_ldt == NULL)
142					panic("could not copy LDT");
143				pcb1->pcb_ldt = pcb_ldt;
144				set_user_ldt(pcb1);
145				user_ldt_free(pcb1);
146			}
147		}
148		return;
149	}
150
151	/* Ensure that p1's pcb is up to date. */
152#ifdef DEV_NPX
153	if (td1 == curthread)
154		td1->td_pcb->pcb_gs = rgs();
155	savecrit = critical_enter();
156	if (PCPU_GET(npxthread) == td1)
157		npxsave(&td1->td_pcb->pcb_save);
158	critical_exit(savecrit);
159#endif
160
161	/* Point the pcb to the top of the stack */
162	pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
163	td2->td_pcb = pcb2;
164
165	/* Copy p1's pcb. */
166	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
167
168	/*
169	 * Create a new fresh stack for the new process.
170	 * Copy the trap frame for the return to user mode as if from a
171	 * syscall.  This copies most of the user mode register values.
172	 */
173	td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
174	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
175
176	td2->td_frame->tf_eax = 0;		/* Child returns zero */
177	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
178	td2->td_frame->tf_edx = 1;
179
180	/*
181	 * Set registers for trampoline to user mode.  Leave space for the
182	 * return address on stack.  These are the kernel mode register values.
183	 */
184	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
185	pcb2->pcb_edi = 0;
186	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
187	pcb2->pcb_ebp = 0;
188	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
189	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
190	pcb2->pcb_eip = (int)fork_trampoline;
191	/*-
192	 * pcb2->pcb_dr*:	cloned above.
193	 * pcb2->pcb_ldt:	duplicated below, if necessary.
194	 * pcb2->pcb_savefpu:	cloned above.
195	 * pcb2->pcb_flags:	cloned above.
196	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
197	 * pcb2->pcb_gs:	cloned above.
198	 * pcb2->pcb_ext:	cleared below.
199	 */
200
201	/*
202	 * XXX don't copy the i/o pages.  this should probably be fixed.
203	 */
204	pcb2->pcb_ext = 0;
205
206        /* Copy the LDT, if necessary. */
207	mtx_lock_spin(&sched_lock);
208        if (pcb2->pcb_ldt != 0) {
209		if (flags & RFMEM) {
210			pcb2->pcb_ldt->ldt_refcnt++;
211		} else {
212			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
213				pcb2->pcb_ldt->ldt_len);
214			if (pcb2->pcb_ldt == NULL)
215				panic("could not copy LDT");
216		}
217        }
218	mtx_unlock_spin(&sched_lock);
219
220	/*
221	 * Now, cpu_switch() can schedule the new process.
222	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
223	 * containing the return address when exiting cpu_switch.
224	 * This will normally be to fork_trampoline(), which will have
225	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
226	 * will set up a stack to call fork_return(p, frame); to complete
227	 * the return to user-mode.
228	 */
229}
230
231/*
232 * Intercept the return address from a freshly forked process that has NOT
233 * been scheduled yet.
234 *
235 * This is needed to make kernel threads stay in kernel mode.
236 */
237void
238cpu_set_fork_handler(td, func, arg)
239	struct thread *td;
240	void (*func) __P((void *));
241	void *arg;
242{
243	/*
244	 * Note that the trap frame follows the args, so the function
245	 * is really called like this:  func(arg, frame);
246	 */
247	td->td_pcb->pcb_esi = (int) func;	/* function */
248	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
249}
250
251void
252cpu_exit(td)
253	register struct thread *td;
254{
255	struct pcb *pcb = td->td_pcb;
256
257#ifdef DEV_NPX
258	npxexit(td);
259#endif
260	if (pcb->pcb_ext != 0) {
261	        /*
262		 * XXX do we need to move the TSS off the allocated pages
263		 * before freeing them?  (not done here)
264		 */
265		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
266		    ctob(IOPAGES + 1));
267		pcb->pcb_ext = 0;
268	}
269	if (pcb->pcb_ldt)
270		user_ldt_free(pcb);
271        if (pcb->pcb_flags & PCB_DBREGS) {
272                /*
273                 * disable all hardware breakpoints
274                 */
275                reset_dbregs();
276                pcb->pcb_flags &= ~PCB_DBREGS;
277        }
278}
279
280void
281cpu_wait(p)
282	struct proc *p;
283{
284}
285
286/*
287 * Dump the machine specific header information at the start of a core dump.
288 */
289int
290cpu_coredump(td, vp, cred)
291	struct thread *td;
292	struct vnode *vp;
293	struct ucred *cred;
294{
295	struct proc *p = td->td_proc;
296	int error;
297	caddr_t tempuser;
298
299	tempuser = malloc(ctob(UAREA_PAGES + KSTACK_PAGES), M_TEMP, M_WAITOK | M_ZERO);
300	if (!tempuser)
301		return EINVAL;
302
303	bcopy(p->p_uarea, tempuser, sizeof(struct user));
304#if 0		/* XXXKSE - broken, fixme!!!!! td_frame is in kstack! */
305	bcopy(td->td_frame,
306	      tempuser + ((caddr_t) td->td_frame - (caddr_t) p->p_uarea),
307	      sizeof(struct trapframe));
308#endif
309
310	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
311			ctob(UAREA_PAGES + KSTACK_PAGES),
312			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
313
314	free(tempuser, M_TEMP);
315
316	return error;
317}
318
319/*
320 * Convert kernel VA to physical address
321 */
322u_long
323kvtop(void *addr)
324{
325	vm_offset_t va;
326
327	va = pmap_kextract((vm_offset_t)addr);
328	if (va == 0)
329		panic("kvtop: zero page frame");
330	return((int)va);
331}
332
333/*
334 * Map an IO request into kernel virtual address space.
335 *
336 * All requests are (re)mapped into kernel VA space.
337 * Notice that we use b_bufsize for the size of the buffer
338 * to be mapped.  b_bcount might be modified by the driver.
339 */
340void
341vmapbuf(bp)
342	register struct buf *bp;
343{
344	register caddr_t addr, v, kva;
345	vm_offset_t pa;
346
347	GIANT_REQUIRED;
348
349	if ((bp->b_flags & B_PHYS) == 0)
350		panic("vmapbuf");
351
352	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
353	    addr < bp->b_data + bp->b_bufsize;
354	    addr += PAGE_SIZE, v += PAGE_SIZE) {
355		/*
356		 * Do the vm_fault if needed; do the copy-on-write thing
357		 * when reading stuff off device into memory.
358		 */
359		vm_fault_quick(addr,
360			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
361		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
362		if (pa == 0)
363			panic("vmapbuf: page not present");
364		vm_page_hold(PHYS_TO_VM_PAGE(pa));
365		pmap_kenter((vm_offset_t) v, pa);
366	}
367
368	kva = bp->b_saveaddr;
369	bp->b_saveaddr = bp->b_data;
370	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
371}
372
373/*
374 * Free the io map PTEs associated with this IO operation.
375 * We also invalidate the TLB entries and restore the original b_addr.
376 */
377void
378vunmapbuf(bp)
379	register struct buf *bp;
380{
381	register caddr_t addr;
382	vm_offset_t pa;
383
384	GIANT_REQUIRED;
385
386	if ((bp->b_flags & B_PHYS) == 0)
387		panic("vunmapbuf");
388
389	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
390	    addr < bp->b_data + bp->b_bufsize;
391	    addr += PAGE_SIZE) {
392		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
393		pmap_kremove((vm_offset_t) addr);
394		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
395	}
396
397	bp->b_data = bp->b_saveaddr;
398}
399
400/*
401 * Force reset the processor by invalidating the entire address space!
402 */
403
404#ifdef SMP
405static void
406cpu_reset_proxy()
407{
408
409	cpu_reset_proxy_active = 1;
410	while (cpu_reset_proxy_active == 1)
411		;	 /* Wait for other cpu to see that we've started */
412	stop_cpus((1<<cpu_reset_proxyid));
413	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
414	DELAY(1000000);
415	cpu_reset_real();
416}
417#endif
418
419void
420cpu_reset()
421{
422#ifdef SMP
423	if (smp_active == 0) {
424		cpu_reset_real();
425		/* NOTREACHED */
426	} else {
427
428		u_int map;
429		int cnt;
430		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
431
432		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
433
434		if (map != 0) {
435			printf("cpu_reset: Stopping other CPUs\n");
436			stop_cpus(map);		/* Stop all other CPUs */
437		}
438
439		if (PCPU_GET(cpuid) == 0) {
440			DELAY(1000000);
441			cpu_reset_real();
442			/* NOTREACHED */
443		} else {
444			/* We are not BSP (CPU #0) */
445
446			cpu_reset_proxyid = PCPU_GET(cpuid);
447			cpustop_restartfunc = cpu_reset_proxy;
448			cpu_reset_proxy_active = 0;
449			printf("cpu_reset: Restarting BSP\n");
450			started_cpus = (1<<0);		/* Restart CPU #0 */
451
452			cnt = 0;
453			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
454				cnt++;	/* Wait for BSP to announce restart */
455			if (cpu_reset_proxy_active == 0)
456				printf("cpu_reset: Failed to restart BSP\n");
457			enable_intr();
458			cpu_reset_proxy_active = 2;
459
460			while (1);
461			/* NOTREACHED */
462		}
463	}
464#else
465	cpu_reset_real();
466#endif
467}
468
469static void
470cpu_reset_real()
471{
472
473#ifdef PC98
474	/*
475	 * Attempt to do a CPU reset via CPU reset port.
476	 */
477	disable_intr();
478	if ((inb(0x35) & 0xa0) != 0xa0) {
479		outb(0x37, 0x0f);		/* SHUT0 = 0. */
480		outb(0x37, 0x0b);		/* SHUT1 = 0. */
481	}
482	outb(0xf0, 0x00);		/* Reset. */
483#else
484	/*
485	 * Attempt to do a CPU reset via the keyboard controller,
486	 * do not turn of the GateA20, as any machine that fails
487	 * to do the reset here would then end up in no man's land.
488	 */
489
490#if !defined(BROKEN_KEYBOARD_RESET)
491	outb(IO_KBD + 4, 0xFE);
492	DELAY(500000);	/* wait 0.5 sec to see if that did it */
493	printf("Keyboard reset did not work, attempting CPU shutdown\n");
494	DELAY(1000000);	/* wait 1 sec for printf to complete */
495#endif
496#endif /* PC98 */
497	/* force a shutdown by unmapping entire address space ! */
498	bzero((caddr_t) PTD, PAGE_SIZE);
499
500	/* "good night, sweet prince .... <THUNK!>" */
501	invltlb();
502	/* NOTREACHED */
503	while(1);
504}
505
506int
507grow_stack(p, sp)
508	struct proc *p;
509	u_int sp;
510{
511	int rv;
512
513	rv = vm_map_growstack (p, sp);
514	if (rv != KERN_SUCCESS)
515		return (0);
516
517	return (1);
518}
519
520/*
521 * Software interrupt handler for queued VM system processing.
522 */
523void
524swi_vm(void *dummy)
525{
526	if (busdma_swi_pending != 0)
527		busdma_swi();
528}
529
530/*
531 * Tell whether this address is in some physical memory region.
532 * Currently used by the kernel coredump code in order to avoid
533 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
534 * or other unpredictable behaviour.
535 */
536
537int
538is_physical_memory(addr)
539	vm_offset_t addr;
540{
541
542#ifdef DEV_ISA
543	/* The ISA ``memory hole''. */
544	if (addr >= 0xa0000 && addr < 0x100000)
545		return 0;
546#endif
547
548	/*
549	 * stuff other tests for known memory-mapped devices (PCI?)
550	 * here
551	 */
552
553	return 1;
554}
555